Deep learning
Normalization layer
A normalization layer rescales activations using statistics over selected dimensions.
How it works
Normalization keeps signal scales controlled and changes optimization geometry. Layer normalization is common in transformers because it works independently per sequence item.
A concrete example
Illustrative: values [2, 4, 6] become approximately [-1.22, 0, 1.22] before learned scale and shift.
Do not confuse it with Data normalization
a layer normalizes internal activations; data normalization preprocesses inputs.